Last week I had the pleasure of speaking at Evanta’s 2019 CISO Executive Summit in St. Louis. My talk was titled “AI in 2020: What’s Real, What’s Fake and What’s Next?”
As we wrap up 2019, it feels like we are heading into an inflection point with AI. It’s no longer the sexy just-out-of-reach buzz word. All our phones use AI to recognize our voices and faces. A quarter of homes have smart speakers. Many companies – Maritz Motivation included – use AI every day in their business processes. AI is here.
In my session, I highlighted a few key areas which will be important for AI in 2020. Below are two of those.
The human brain is highly sensitive to faces. We can instinctively tell if something looks just a little bit off. Think Carrie Fischer’s posthumous appearance in Rogue One or “young” Robert DeNiro in The Irishman. The computer-generated faces look almost perfect, but you can’t keep your brain from thinking, “This doesn’t look quite right.”
Our brains are so keyed into faces that it was an open question if AI would ever advance to the level where it could fool humans. Similar questions were once raised about Chess, Jeopardy!, and Go, but just like all of those, this is a barrier that AI will hurdle.
DeepFakes are an innovation where an AI maps one face onto another. This technology has gotten incredibly realistic very quickly. It is being used in many humorous and clever applications, see this and this, but the risk of nefarious use is incredibly high. Researchers are trying to find ways to detect fake videos, but it is unclear right now if the mousetraps will be able to stay ahead of the mice.
If you don’t think that AI faces will be able to fool you, I encourage you to check out thispersondoesnotexist.com. Every time you reload the page, the AI will generate a new face. These are often so good that when I show this site to people, they don’t believe these are being generated on the spot. The best way to prove this to yourself is to reload it a few times until you get a weird one. Look specifically at eyeglasses, jewelry, and the backgrounds.
In the past year, there have been big leaps in AI’s usefulness in understanding and generating language.
Doing business internationally is now more accessible than ever. Ebay recently used AI translation to boost sales in Latin America by over 10 percent. That’s a great use case because the item descriptions are user-generated, so someone who doesn’t speak a lick of Spanish can still sell effectively to Spanish-speaking audiences.
The way most translation systems work is by transforming the sounds to text, then translating the text into a different language. That won’t always be the case. Real human speech is more than knowing that “agua” means “water” – the tone of voice is an important element too. Google has developed an end-to-end speech-to-speech translation model which will not rely on the intermediate step of translating to text. This means that the listener will be able to hear your tone of voice in a different language, so they could understand things like sarcasm or urgency.
Not only are the translation models themselves changing, the delivery mechanisms are too. The dream of the Star Trek universal translator, which automatically translates any language in your ear, is closer than you may think. There are in-ear translators on the market now, and in the near future it is easy to imagine that all earbuds that connect to your phone will have this capability natively.
Text generation is another area of language which has gotten exponentially better this year. You are probably used to one-word suggestive text on your smartphone. Now, AI can suggest entire paragraphs. At the conference, we used an AI to create paragraphs in real time, based on prompts about Star Wars and the St. Louis Blues. The audience audibly gasped at how human the results sounded. Give it a shot yourself.
The future is here, let’s make it a good one.
Whenever I talk about AI at a conference, I would say that the majority of innovations I highlight are met with a negative reaction by the audience. I can see the fear of the unknown in their body language, even among technical audiences. I think that’s a natural response and I don’t fault anyone for it – the world is changing without our consent. DeepFakes could be used to influence a political election and text-generators could spread more fake news than fact-checkers could keep up with.
This same pattern of skepticism and fear happened around every major technological innovation in history, from electricity to automobiles to television or the Internet. Just like AI, any of those innovations could be used for harm, but to deny the benefits would be shortsighted. It is our responsibility as ethical business leaders to use AI to make lives better and be the beacons for the positive use of these new technologies. There will always be bad actors, but let them be relegated to the dark corners rather than legitimate businesses. For the rest of us, AI will take another step towards the mainstream in 2020.